Marcel Stanitzki 1 The MAPS ECAL Status and prospects Fermilab 10/Apr/2007 Y. Mikami, O. Miller, V. Rajovic, N.K. Watson, J.A. Wilson University of Birmingham J.A. Ballin, P.D. Dauncey, A.-M. Magnan, M. Noy Imperial College London J.P. Crooks, M. Stanitzki, K.D. Stefanov, R. Turchetta, M. Tyndel, E.G. Villani Rutherford Appleton Laboratory

Marcel Stanitzki 2 Overview Introduction Status  Sensor Design  Sensor Simulation  DAQ/Testing  Detector Simulations Next steps

Marcel Stanitzki 3 Introduction This is done within the context of CALICE Not for a particular detector concept Development of an alternative readout sensor for a SiW based ECAL “Swap-In” Solution leaving mechanical structure untouched Using MAPS with high granularity and digital readout  Should help Particle Flow Algorithms  But it will be a Tera-Pixel Calorimeter...

Marcel Stanitzki 4 What are MAPS ? Monolithic Active Pixel Sensors Integration of Sensor and Readout Electronics Manufactured in Standard CMOS process Collects charge mainly by diffusion Development started in the mid-nineties, now a mature technology

Marcel Stanitzki 5 Hybrid Pixels and MAPS Pixels Bump Bonds Readout Electronics LHC-style Hybrid Pixel sensor MAPS

Marcel Stanitzki 6 MAPS in Detail Incoming particle Diode s Epi-Layer (up to 20 µm) CMOS Wafer (~300 µm) Electronic s MAPS architecture: Sensor and the electronics are integrated in one wafer Charge Collection mainly in epi-layer Charge collected mostly due to diffusion

Marcel Stanitzki 7 The ECAL MAPS Pixel Size (50 x 50 µm) Binary Readout (1 bit ADC realized as Comparator) 4 Diodes for Charge Collection Time Stamping with 13 bits (8192 bunches) Hit buffering for entire bunch train Capability to mask individual pixels Threshold adjustment for each pixel For the MAPS ECAL a specific MAPS was designed:

Marcel Stanitzki 8 A new process technology Simulation showed, that the electronics n-wells absorb a lot of charge (affects the signal) We isolated the n-well with a “deep p-well” implant (3 µm thick) Standard for deep n-well (“triple well”) Novel INMAPS process used for the ECAL MAPS Incoming particle Diodes Epi CMOS Wafer Electronics Deep p- well

Marcel Stanitzki 9 Sensor Electronics Two types of pixel readout Shaper & Sample Deadtime (~600 ns/ 450 ns ) Simulation shows similar noise characteristics Both share the Comparator design and everything downstream Having two front-end architectures allows us to explore several ideas at once

Marcel Stanitzki 10 The two pixels Rst Vrst Preamp PreRst Buffer s.f Cfb Cin Buffer s.f Vth+ Vth- Reset Sample Cstore --ns Preamp Shaper Rst Cpre Cfb Rfb Rin Cin Vth+ Vth- Pre- Shaper Pre- Sampler

Marcel Stanitzki 11 The Pixel 50x50 µm size 0.18 µm process 12 µm Epi-layer (for test run) Deep p-wells 6 metal layers V transistors V transistor 36 capacitors

Marcel Stanitzki 12 The sensor unit (V1.0) Consists of 42x84 pixels Has a logic strip for  5 pixels wide  Hit buffering using SRAM technology, 19 Hits per Row  Time stamping (13 bit)  Configuration registers  the only part with Clock lines Logic strip is a “dead area” for particle detection (~ 11 % inefficiency) Pixel s Logic Strip 42 pixels 5 pixels 84 pixels

Marcel Stanitzki 13 Data format A row of 42 pixels is split into 7 groups of 6 pixels each (“patterns”) The logic writes the following data format for each row  Time stamp (13 bits)  pattern number (3bits)  pattern (6 Bits) 1 Hit = 22 Bits On top :Row Enconding (9 Bits) 1 Hit = 31 Bit altogether

Marcel Stanitzki 14 The test sensor (V1.0) Pre-Sample Architectur e Pre-Shaper Architectur e 8 units (1 x 1 cm) A unit is uniform, but units are different 2 pixel architectures 2 capacitor arrangements 6 million transistors in total Capacitor V1 Capacitor V2

Marcel Stanitzki 15 Sensor Simulation We are using Centaurus TCAD to simulate the sensor Using CADENCE GDS file for pixel description Simulate diodes from adjacent pixels for charge sharing effects Detailed Pixel performance studies  Collection Efficiency  Charge Collection Time  Signal/Noise

Marcel Stanitzki 16 The setup Bias: n-Well 1.8/1V Diodes: 1.5V Diode s Adjacent Diodes Electronic s Substrat e (left floating) Epi- Layer

Marcel Stanitzki 17 Simulation Setup 7 Hit Points simulated 1 µm distance

Marcel Stanitzki 18 Charge Collection Main parameter to vary is Diode SizeILC Bunch spacing ~ 300 ns Total charge generated : ~ 1300 electrons

Marcel Stanitzki 19 Signal/Noise *N.B. S/N 0.9µm N = 16 e - *N.B. S/N 1.8µm N = 16.5 e - *N.B. S/N 3.6µm N = 21.3 e - ● Signal to Noise > 15 for 1.8 µm Diode Size ● Some uncertainty for the absolute Noise levels, due to simulation imperfections ● Critical Measurement with the real sensor

Marcel Stanitzki 20 Noise Occupancy Noise for 2880 bunches With Noise=O(10 -6 )  P=0.3 % for 1 hit per pixel  P= % for 2 hit per pixel But O(10 12 ) pixels !  ~ single hits  ~ double hits  ~0 triple hits Per Row (42 pixels) 0.15 Hits

Marcel Stanitzki 21 MAPS DAQ & Testing Development of DAQ board and firmware has started Complete test setup foreseen  Cosmics  Sources  Laser  Test beam

Marcel Stanitzki 22 RAL Laser Test setup Powerful Laser setup 1064, 532 and 355 nm Wavelength Accurate focusing (<2 µm at longest wavelength) Pulse Width 4 ns 50 Hz Repetition rate Fully automatized Will be used to test the MAPS

Marcel Stanitzki 23 Simulation Chain Event Generation e.g. Pythia Event Generation e.g. Pythia Detector Simulation MOKKA Detector Simulation MOKKA MAPS Digitizer as MARLIN processor LCIO Detector Simulation SLIC LCIO

Marcel Stanitzki 24 Mokka Detector Simulation Implementation of the MAPS into MOKKA  Patched MOKKA x50 µm pixel size 15 µm “Active Area” (Epi-layer) Detector Model used LDC01(Sc) ECAL with 30 layers  20 layers 2.1 mm Tungsten  10 layers 4.2 mm Tungsten Charge diffusion and thresholds are implemented in a separate “Digitization” step

Marcel Stanitzki 25 Shower Shapes GEANT4 : One 20 GeV Electron shot along y-axis

Marcel Stanitzki 26 Simulation Chain Event Generation e.g. Pythia Event Generation e.g. Pythia Detector Simulation MOKKA Detector Simulation MOKKA MAPS Digitizer as MARLIN processor LCIO Detector Simulation SLIC LCIO

Marcel Stanitzki 27 Running with SLIC MAPS 50 μm  50 μm p

Marcel Stanitzki 28 Simulation Chain Event Generation e.g. Pythia Event Generation e.g. Pythia Detector Simulation MOKKA Detector Simulation MOKKA MAPS Digitizer as MARLIN processor LCIO Detector Simulation SLIC LCIO

Marcel Stanitzki 29 Charge sharing algorithm Geant4 E init in 5x5 μm² cells Sum energy in 50x50 μm² cells E sum Apply charge spread E after charge spread Add noise to signal hits with σ = 75 eV + noise only hits prob  ~ 10 6 hits in the whole detector BUT in a 1.5 x 1.5 cm tower : ~30 hit in 30 layers. %E ini t E init Register the position and the number of hits above threshold

Marcel Stanitzki 30 Basic Hit Clustering Hit Clustering Loop over hits classified by number of neighbors Number of neighbors < 8 : count only 1 (or 2 for last 10 layers) and discard the neighbors 8 neighbors AND one of the neighbor has 8 neighbors : count 2 (or 4) and discard the neighbors 3 or 4

Marcel Stanitzki 31 Clustering (II)

Marcel Stanitzki 32 Beam background Done using GuineaPIG Trying to estimate beam induced background in the ECAL Testing two scenarios  500 GeV Baseline  1 TeV High Lum 1 TeV High Lum is “worst-case” scenario

Marcel Stanitzki 33 1 TeV High Luminosity “Ring of Fire” for small ECAL Radii

Marcel Stanitzki 34 Beam Background Occupancy MAPS Reset time

Marcel Stanitzki 35 Design Issues Pixel parameters  Pixel size  Number of Diodes / Diode size PCB/Readout Chips Stave Structure Power DAQ Manufacturing

Marcel Stanitzki 36 Pixel parameters MAPS Pixels improve spatial resolution/granularity by a factor of ~ 1000 compared to analog pad ECAL Lower pixel size is set by size of the integrated electronics (lower boundary of 50 µm) and charge diffusion Upper bound set by charge collection time/efficiency and multiple hits No fixed upper bound, reasonable value is around 100 µm Best performance found with 4 diodes is for 1.8 µm diode size

Marcel Stanitzki 37 PCB/ Readout Chip All the electronics is integrated within the the sensor No need for  Complicated PCB design  Dedicated Readout Chip Still needs to provide Power/Clocks/Commands to the MAPS Can be done by “Stave Controller” at the end of the Stave

Marcel Stanitzki 38 Stave Structure MAPS can be used as swap-in solution without alterations to the mechanical design (Baseline) One can also take further use of MAPS benefits CALICE

Marcel Stanitzki 39 How it could look like Take advantage of MAPS benefits Lack of hybrids/ASIC allow less complex/thinner PCB Thinner sensors (down to 100 µm) Bump-bond MAPS MAP S Tungsten Stave Controller with optical link PC B

Marcel Stanitzki 40 Power Cooling for the ECAL is a general issue Power Savings due to Duty Cycle (1%) Target Value for baseline ECAL 4 µW/mm 2 (CALICE) Current Consumption of MAPS ECAL: 40 µW/mm 2 depending on pixel architecture Compared to analog pad ECAL  Factor 1000 more Channels  Factor 10 more power Advantage: Heat load is spread evenly

Marcel Stanitzki 41 Power prospects V1.0 not been optimized for power consumption  Proof of Concept and Technology  Not the final product Options to be explored  Larger pixel (50 µm->100 µm) Factor 4 less  Longer integrations times if pile-up acceptable, possible factor of 2  Smaller feature size (~30-50 %)  Lowering Operating Voltages ( ~10%) Sensor V1.0 will allow us to explore some of these

Marcel Stanitzki 42 DAQ data volume Physics rate is not the limiting factor Beam background and Noise will dominate Assuming 2880 bunches and 32 bits per Hit  10 6 Noise hits per bunch  ~O(1000) Hits from Beam background per bunch (estimated) Per bunch train  ~88 Gbit / 11 Gigabyte  Readout speed required 440 Gbit/s  CDF SVX-II can do 144 Gbit/s already

Marcel Stanitzki 43 Manufacturing Sensor manufacturing  Need for large scale process ( m 2 )  Factor 10 of CMS (205 m 2 )  CMOS is an industry standard process  Many foundries can do it  CMOS wafers are readily available  CMOS is ~2 cheaper than “HEP-style” silicon Stave manufacturing  Less complex structure due to lack of VFE ASIC  No substrate connection to ground required

Marcel Stanitzki 44 What happens next ? Submit Sensor V1.0 April 23 rd Sensor V1.0 due back Mid July Improve/enhance GEANT simulation Testing Sensor V1.0 Do physics studies with a MAPS based ECAL Improve sensor simulation with data from V1.0

Marcel Stanitzki 45 Sensor V2.0 Will be based on experience made with V1.0 Larger structures 2x2cm per module Only one Pixel readout architecture More power optimized Submission date Summer 2008

Marcel Stanitzki 46 Summary MAPS effort is advancing well Sensor V1.0 is almost done and be submitted next week Will be the proof of concept Simulation of a MAPS ECAL is in place  Will need tuning with results from sensor V1.0 Sensor V2.0 will be close to real system Will be ready when Detector EDR will be required.

Marcel Stanitzki 47 Backup

Marcel Stanitzki 48 Diode placement Classical problem  Place n circles in a square  No analytical solution Only 4 Diodes as a starter Mathematics faces reality  Constraints due to Design Rules  Electronics  Space Numerical Solution 50 µm Charge Collection Efficiency

Marcel Stanitzki 49 Spatial resolution in x Moliere Radius Impact of B field GEANT4 :20 GeV Electrons shot along y-axis

Marcel Stanitzki 50 Linear Response Linear response for electrons up to 400 GeV GEANT4 level without charge diffusion

Marcel Stanitzki 51 Resolution Good resolution over wide energy range GEANT4 level without charge diffusion

Marcel Stanitzki 52 First results Algorithm depends on accurate simulation input from Centaurus First results shows algorithm work nicely Does not take into account deep p-well yet Will be updated with the latest pixel simulations and noise estimates

Marcel Stanitzki 53 Hit weighting Hit Counting: Hit x 2 in the last 10 layers to account for the double tungsten thickness.

Marcel Stanitzki 54 Dead area We have an area of 5 dead pixels every 42 sensitive pixels for the logic strip. σ(E)/E = α/√E  α = asymptotical value  α = 42 sensitive pixels

Marcel Stanitzki 55 Cooling cont'd

Marcel Stanitzki 56 ILC configurations